Enable OpenCL on macOS

[wuyakuma]

Although OpenCL was marked as deprecated in macOS 10.14, it still seems to work very well.
With some minor changes, OpenCL traversal on macOS can now function, and the performance is fantastic (see the screenshot).

The results were verified by ValidateTraceResult() and also dumped to images.

PS: rayData.hit was not cleaned between ValidateTraceResult() calls. After adding some code to clean the data, I found that gpu4way is broken in the release build. However, this seems to be caused by BVH building not the traversal. When replacing bvh4_gpu->bvh4Data with the debug version, everything works fine. Maybe another macOS release compiler mystery?

[jbikker]

Very impressive work! I can't change it obviously, but I am grateful for this PR.

[jbikker]

Wait a minute @wuyakuma does that screenshot say 16B rays per second? That is insane. :)

[wuyakuma]

Wait a minute @wuyakuma does that screenshot say 16B rays per second? That is insane. :)

Yeah, that's about 10x faster than my Windows machine. Totally unexpected. If I hadn't seen the verification myself, I still wouldn't believe it.

[wuyakuma]

I still have concerns about the results due to the significant performance difference. It would be great if someone could verify it on macOS too.
If anyone interested, here are my local changes to tiny_bvh_speedtest.cpp, which include cleaning raydata between each test and dumping the results to an image (stb_image_write is required).
tiny_bvh_speedtest.cpp.txt